Read the original document by opening this link in a new tab.
Table of Contents
Abstract
1. Introduction
2. Background
3. Infinite Mixture Prototypes (IMP)
3.1. Adapting capacity by learning cluster variance
3.2. Adapting capacity by multi-modal clustering
4. Ablations and Alternatives
Summary
The document discusses the proposal of infinite mixture prototypes for adaptively representing simple and complex data distributions for few-shot learning. By inferring the number of clusters, infinite mixture prototypes interpolate between nearest neighbor and prototypical representations, improving accuracy and robustness. The importance of adaptive capacity and multi-modal clustering are highlighted, showing significant accuracy improvements over existing methods. The paper also extends the approach to semi-supervised and unsupervised clustering. Various methodologies and background concepts for few-shot classification are explained in detail.